Thanks to everyone who read our explanation of political polling. In less than eight hours, it screamed to the top of our all-time most-popular features here at the Big G and got considerable attention from much bigger media outlets.
That said, it attracted some mail. First up, YetAnotherJ wrote in:
Interesting post. But you did make one questionable statement. You said:In 2004, 37% of voters went for Democrats, 37% of voters went for Republicans, and 26% claimed to be independent. But enough independents leaned Republican for GOP candidates to win a few elections that year.
Bush actually lost the independents by 1% (at least according to the exit polls which have their own margin of error issues). Bush won because he was able to pull 5 points more dems than Kerry did GOP. That is not as likely to happen this year. On the other hand, your premise that this year’s winner is likely to be the one who persuades more independents is likely to be correct.
Indeed, and the Czar originally had that explanation in there. However, our numbers reflected all 2004 elections—not just the presidential contest—and we stand by ‘em as quoted.
It would be a massive project, but it would be interesting to take this year’s polls and apply the exit poll party ID split to those polls. You will undoubtedly see an inflection point after the first debate. But that will likely tell us quite a bit about the true course of the election, as seen with 20/20 hindsight.
Another interesting question is why the polls party IDs are so bad this year. If you discount the conspiracy theories that pollsters are all secret democrats trying to give Obama the election, there are two interesting data points and a third anecdotal bit of data that may point to an answer. First, Pew polling is reporting that they are now seeing 9% of interviews being concluded. This is less than about 1/4 what it was just a decade ago. So each completed interview represent about 11 other voters. You can see that just a little skew in your who completes the poll and you have a big divergence from reality. I don’t know if anyone has done a calculation of margin of error given such a low response rate, but I suspect that it may be higher than based just on the numbers polled. The second data point is that trust in the media is setting new lows every year, especially among conservatives. The anecdotal information is that I see a trend on real clear politics poll reports that shows media sponsored polls (e.g. NYT, WaPo) show better numbers for Obama than polls that aren’t identified with a media organization directly (e.g. Gallup or Rassmussen). Again, discounting the conspiracy theory that the MSM is somehow putting their thumb on the scale for Obama (okay, we know they are, but it isn’t clear they are doing this by rigging polls), we can wonder if the two data points don’t play into MSM polls being off. Specifically, If conservatives are just that much more likely to hang up when they hear NYT than liberals. If so, you can see how this would sway things.
Imagine that you are trying to complete 900 interviews. You start by calling 10000 people. If there where 37% dems, 37% GOP and 26% indies (aka 2004), your 900 interviews should reach 333 dems, 333 GOP and 234 indies out of an original pool of 3700 dems, 3700 GOP and 260 indies. Now what if GOP voters where just 1% more likely to hang up than the average. So the dem response rate is 9%, the GOP response rate is 8%. To get the 900 interviews (and make the 9% number balance), you get a 9.61% response by dems and a 9.4% response by indies. So instead of 333 dems, 333 GOP and 234 indies, you get 358 dems, 296 GOP and 246 indies. Assuming both sides where drawing 90% of their base and 10% of the other party’s base and indies were exactly 50-50, you would have expected to see a poll showing a tie. Instead, you get a 53% to 47% poll result on what should have been 50% to 50%. That is because GOP voters were just 1% less likely to answer their poll than the average.
How are pollsters going to handle something like this (assuming they recognize the “Dewey defeats Truman” error in their polling methods)? One step would be to not identify the poll with a MSM, which would make it slightly less likely to get a GOP hang up. Another is to consider weighting based on party ID and “voter intensity”. They could also release a range of results based on probably voter turnout (e.g. their best guess of voter turn out) and variations up and down from there. If nothing else, it would allow you to better compare polls as apples to apples as you are looking at a D+6 poll and a R+1 poll and wondering how they can be so far apart.
Anyway, I enjoyed your article.
Thanks! Indeed, you bring up a very important element to political polls that most folks don’t realize: all the poll data you see on television, newspapers, the Web, etc., all stem from about a 9% participation rate. That’s right: all this fuss is being made by less than one in ten. That’s decimation, by the by.
Also, Island Dweller tees me up (and off!) with a perfect setup:
Your august majesty:
I am wondering how the accuracy of the unceasing polls we have been treated to (or have had inflicted on us) in the recent past will reflect on print newspaper circulation. As you are undoubtedly aware, due to your omniscience, print newspaper circulation, including numerous “journals of record,” has been in a downward death spiral for some time now. If these major rags – excuse me, newspapers – are proven wrong tomorrow and Mr. Romney comes out on top, one has to wonder why a subscriber would want to continue to receive such nonsense. One more nail in the coffin representing the demise of the print media. BTW it’s not just newspapers – have you seen how small – and thin – “Newsweek” and “Time” have gotten over the past few years? I know the internet has a lot to do with this, and a proportion of the misguided and misled will continue to seek this stuff out, no matter in what form it’s presented – but the thinness in the aforementioned magazines represents lost sponsorship that is now appearing on the web, along with lost readers who have finally realized they were being misled, or could get it cheaper and much faster on the internet. Skewing election coverage is another sure way to lose circulation, it would seem.
Your awed and obedient correspondent and minion,
There will indeed be a reckoning, even if Obama manages a win. Neither side can be very pleased with the wholly inaccurate polling information going on right now, and because campaigns pay those bills, they will insist on more accurate methods going forward.
But your question is a bit different: what is the effect on the dying newsprint media? Frankly, they’re already dead; what arrives in your driveway—assuming you even still subscribe—is a ghostly echo. Time is struggling to survive; Newsweek, we are happy to report, ceases physical publication on December 31, 2012. The venerable Chicago Tribune is now a smaller formatted, multi-page pamphlet wrapped around sale papers, as are many other publications. The Freep is already gone and is now an unusual website. You note this yourself.
When Romney wins, they will not be held liable for bad reporting—they will blame the pollsters. Anyone who still subscribes to an actual, physical newspaper anymore might be upset with the juvenile and unprofessional media coverage, but they would be beating a very dead horse.
The end of media bias in general will take a few more years, but as alternative media increasingly presents pro-conservative, pro-Republican, and pro-libertarian coverage, you will see this fall apart. Television now embraces it: Fox is openly right-wing, MSNBC and its twelve viewers are blatantly left-wing, and CNN is strongly considering a tilt to either direction (the Czar hears a rightward tilt based on Fox’s higher ratings and MSNBC’s negligible ratings).
Stand by. We’ll see.